Catching a Field Trial Going Wrong Mid Pass Scoring a Grower's Appetite for Risk, Then Steering the Planter Back Into Compliance

This application covers a computerised way of running agricultural experiments on working farms rather than on research plots. A central system scores thousands of fields for how much risk their manager will tolerate, invites the best candidates to take part in a trial, works out where on each field the test plots should sit, then watches the planter or sprayer in real time and intervenes when the machine departs from the trial design. It was filed by Climate LLC, the digital farming business behind Climate FieldView, and names Thomas Gene Ruff, Jason Kendrick Bull, Nicholas Charles Cizek, Brandon Rinkenberger and Doug Sauder as inventors. The application is a divisional of Australian application 2024204214.

The Problem

Agricultural advice is easy to give and hard to prove. The background section puts the difficulty in two sentences worth paraphrasing. A field manager who is confident about the results of their own practices has no particular reason to trust a recommendation to change them, and even a manager who does agree to change cannot afterwards tell whether the improvement came from the new seed, the new fertiliser rate, or a kind season. Without that separation, the specification says, the manager cannot decide whether to repeat the practice next year.

The classical answer is the agricultural experiment station, where treatments are laid out in small replicated plots with randomised assignment, hand planted if need be, and the statistics come out clean. The answer scales badly. A station plot is a few metres across and sits on ground that may not resemble anyone’s paddock, and a result from one soil type in one season is a weak guide to a farm two hundred kilometres away.

The alternative is the on-farm strip trial, where a grower runs a strip of the new treatment down the length of a paddock beside a strip of their normal practice and compares the yield monitor traces at harvest. That is realistic, but it is statistically fragile. A single treated strip against a single control strip is one replicate, and paddock yield varies enough between one end and the other that the difference between the strips may be entirely soil. The fix in theory is blocking and replication, meaning several matched pairs scattered through comparable ground with the treatment randomly assigned within each pair. The fix in practice runs into the machinery. A planter is twelve or twenty-four rows wide, it plants at speed, it turns on headlands, and switching seed variety or rate mid pass is either impossible or slow. A grower planting three hundred hectares in a narrow weather window will not stop to re-read a trial map, and a strip planted in the wrong place or at the wrong rate silently destroys the replicate without anyone noticing until harvest.

The specification also frames a commercial problem underneath the statistical one. Someone has to bear the risk of the strip that yields less, and growers with good reason to be conservative are the ones least likely to volunteer.

What This Invention Does

Claim 1 is a method for implementing trials in agricultural fields, performed by what the document calls an agricultural intelligence computing system. It is long, and the length is the point, because almost every step is a limitation on how the trial is chosen, placed, policed and paid for.

The system receives field data for many fields and identifies target fields based on a risk tolerance, and the claim does not leave risk tolerance as a vague notion. It writes the formula into the claim itself, in two alternative forms:

Rt = S + N + Ex + D + Y + Eq + M, or Rt = S N Ex D Y Eq M * R0

where S is the existence of particular traits in the seed being planted, N is the percentage of the field under a new product, Ex is the number of identified experiments on the field, D is the difference in seeding population between the field and the regional average, Y is the predictability of yield variability, Eq is the presence of particular types of equipment, M is references to risky activities on social media, and R0 is a base risk rate. The body explains what each term is meant to capture. A manager who plants few hybrids, plants varieties bred for high yield under optimal conditions, habitually under-applies pest control relative to best practice, runs variable rate gear or active downforce management on the planter, or talks about riskier activities online, is read as a manager who will take a chance on a test strip.

From there the claim tracks the whole life of the trial. The system sends a trial participation request to the field manager’s device and waits for acceptance, so nothing happens without consent. It then determines locations for the trial on the basis that those locations sit within at least one management zone of the field, which is the claim’s way of saying that treatment and control must be compared on ground that is agronomically alike. It sends the locations and implementation instructions out.

The distinctive part is what happens next. The system receives application data while the agricultural implement is performing the activity, and at least some of that data comes from field sensors or from sensors built into the implement itself, as the work is being done. It compares that stream against the rules of the trial, determines that the field is out of compliance while the machine is still working, identifies additional locations inside the same management zone that would restore compliance, displays a warning on the manager’s screen naming those locations, directs the implement to them, and then automatically controls at least one operating parameter of the implement through an application controller so that the activity is performed correctly at the new locations. Finally it receives result data and computes a benefit value for the trial.

In plain terms, the claim is a trial that repairs itself mid pass. If the planter drops the wrong population across the strip that was supposed to be the control, the system does not wait until harvest to discover a ruined replicate. It finds equivalent ground still ahead of the machine, tells the driver, steers the machine there and sets the rate itself.

The worked numbers in the body are modest but concrete. A trial might require a testing location planted at 35,000 seeds per acre, and the plot changes colour on the manager’s map once the system confirms that rate was actually delivered. A manager may say through the interface that they will give up five per cent of the field, and the system computes the number of test plots from the field area, that percentage, and the plot area. Alternatively the manager states a minimum treatment effect they want to be able to detect, in bushels per acre, with a given signal to noise ratio, and the number of plots is computed from that, the standard deviation of yield difference between candidate plots, and the target effect size. Plot size and shape come from historic yield variability: rapidly fluctuating yield calls for larger plots, long wavelength trends call for smaller ones, and the chosen configuration is the one that minimises total testing area for an acceptable level of statistical significance.

On the commercial side, the body gives an example of the arithmetic behind a guarantee. If the system predicts a practice worth 20 bushels per acre and the crop is around four dollars a bushel, the expected gain is eighty dollars an acre, so a system configured to ask for ten per cent of expected profit sends an offer guaranteeing a fifteen bushel lift at a cost of eight dollars an acre. Elsewhere it offers free seed in exchange for ten per cent of the revenue increase if the lift exceeds 20 bushels per acre, and rebates graded in five bushel steps if it falls short. Another passage sets the responsiveness threshold used to pick fields at a 1.5 bushel yield lift for every extra 1,000 seeds, against an average slope of about 2.8 on the density and yield curve for a hybrid.

Key Features

• Risk tolerance as a claimed formula. The claim recites two alternative expressions, additive and multiplicative, with seven named terms plus a base risk rate, rather than leaving the selection of trial candidates to unspecified modelling.
• Social media as an input variable. The term M is defined in the claim as references to risky activities on social media, which is an unusually direct statement that a grower’s public posting behaviour is a feature in the field selection model.
• Placement constrained to management zones. Trial plots and replacement plots must sit within at least one management zone of the field, so treatment and control are compared on ground the system already regards as agronomically comparable.
• Compliance checked from implement sensors during the pass. A subset of the application data must come from field sensors or sensors integrated with the implement while the activity is being performed, not from a report filed afterwards.
• Automatic relocation and machine control. On detecting non compliance the system identifies additional locations, warns the operator, directs the implement there and sets at least one operating parameter through an application controller.
• Plot geometry derived from yield variability. The body computes optimal plot size, shape and number from historic or modelled yield variability, buffer areas included, minimising total testing area for a required significance level.
• A benefit value computed from the result. The last step of the claim turns harvest data into a number, which is what the specification’s guarantee and rebate structures are priced against.

Who Is Behind It

The applicant is Climate LLC, the business most people know as The Climate Corporation. It began life as WeatherBill in 2006 selling weather insurance to farmers, was bought by Monsanto in 2013 for a reported 1.1 billion dollars, and passed to Bayer when Bayer acquired Monsanto in 2018. The company name on the title page reflects a later change: the corporation became the limited liability company Climate LLC, and today the entity sits inside Bayer Crop Science rather than standing alone. The consumer-facing product is Climate FieldView, a platform that pulls planting and harvest data off machinery, layers satellite imagery and weather over it, and generates variable rate prescriptions. The site currently offers the platform across the Americas, much of Europe and Australia.

The specification carries its own small piece of corporate history. The copyright notice at paragraph 1 reads 2018 The Climate Corporation, so the text predates the rename even though the application was filed in 2026 under the new name. The claim’s preoccupation with downforce, seeding rate and application controllers is consistent with the company’s shape at the time it was drafted: in February 2014 the Climate Corporation was merged with Monsanto’s Integrated Farming Systems and Precision Planting divisions, which is to say with a planter hardware business. The specification says nothing about the five named inventors beyond their names, and none of them has a public profile worth linking here.

On priority, the Australian specification is silent. There is no cross reference paragraph claiming priority or convention status anywhere in the body. The four United States provisional applications cited at paragraph 87, numbers 62/154,207, 62/175,160, 62/198,060 and 62/220,852, are cited as descriptions of weather monitoring apparatus whose disclosure the document assumes, not as priority documents. The only relationship this specification establishes is the title page entry recording it as a divisional of Australian application 2024204214. Because the document states no priority claim, the Priority Country row below records that fact rather than guessing at a chain.

One drafting note for readers who go to the source. Claim 1 refers to “the agricultural implement” at its compliance step without having introduced an agricultural implement earlier in the claim, so the article’s description of the machine is drawn from the body. The claim also tidies up variable names the body does not: the body’s additive formula reuses the letter E for two different terms, while the claim separates them as Ex and Eq.

Why It Matters

There is a real gap between what agronomic research produces and what a farmer can act on. Replicated small plot work gives defensible statistics about ground that is not yours. A neighbour’s strip trial gives you ground you recognise and statistics you should not trust. Everything interesting in precision agriculture for the last twenty years has been an attempt to close that gap by turning ordinary farm machinery into an experimental apparatus, using the fact that a modern planter already knows where it is to the centimetre and already varies its rate on the move.

What this claim adds is the admission that the apparatus will be operated by someone who is busy. The hardest part of an on-farm trial is not designing it, it is getting it executed correctly by a driver working sixteen hour days in a planting window, on a machine that is doing nine other things. Most systems handle that by sending a prescription file and hoping. This one assumes the prescription will be departed from, watches for the departure while there is still field left, and rebuilds the design around the error. Whether that is patentable subject matter is a question for the examiner, but as a description of how field trials actually fail it is accurate.

The risk tolerance formula is the part that will attract attention, and not only because it works. Scoring a farmer on their equipment, their seeding rate relative to the district, their willingness to under-spray and their social media posts, in order to decide who receives an offer, is profiling. It may be entirely benign profiling in service of a trial nobody is obliged to accept. It is still a claim that says, in terms, that what a grower posts online is an input to a commercial targeting decision made by their agronomy software provider. Readers who care about how much of farming’s data exhaust now sits with a handful of platforms will find this claim a clear illustration of what that data is for.

Related Concepts

• Design of experiments – the statistical discipline whose replication and randomisation requirements this system is trying to satisfy on working farms.
• Field experiment – the general category of experiment run in real conditions rather than a controlled setting.
• Replication (statistics) – repeating a treatment several times, which is what the multiple testing locations in each management zone provide.
• Randomized experiment – the random assignment of treatment and control that the body describes applying to trial locations to remove bias.
• Precision agriculture – the broader field of geolocated, variable rate farm management that makes an automated on-farm trial possible at all.

---

AU 2026202178 was published in the Australian Official Journal of Patents on 9 April 2026 and is open for public inspection. Patent applications represent inventions that are sought to be protected and do not necessarily reflect commercially available products.